Real-Time Template Matching Based on Local Stable Pixels

Abstract

In this paper, a template matching method based on local stable pixels is proposed. First, gradient magnitude and orientation difference are utilized to evaluate the local pixel information for local stable pixel detection. Second, an updated strategy based on gradient magnitude standard deviation is designed to detect local stable pixels. Finally, a coarse-to-fine search approach based on the detected stable pixels is employed to acquire rotation angle. Experimental results demonstrate that the proposed method is robust to arbitrary rotation, occlusion, complex background clutter, blur, noise and it is able to meet the demand of real-time processing for visual positioning project.